As the global AI wave continues to surge and demand for data center power and computing capacity repeatedly reaches new highs, the All-Photonics Network (APN) is emerging as a critical infrastructure pillar supporting the next generation of AI. On June 3, 2026, during Computex, the Cloud Computing & IoT Association in Taiwan (CIAT) hosted the "Building AI Infrastructure for IOWN Use Cases" seminar at Taipei 101, drawing more than 200 industry professionals and underscoring the growing importance of all-photonics networks in the era of AI data centers.

The seminar brought together partners from the IOWN Global Forum, including NTT, 1Finity, GigaIO, Saviah Technologies, Pegatron, Chunghwa Telecom, Accton/Edgecore, UfiSpace, and ITRI. Across six exhibition zones, these companies showcased how IOWN's all-photonics network can be used to build distributed AI infrastructure, sharing their vision, technologies, and real-world implementation case studies.

Opening Remarks: Taiwan's Opportunity in the AI Era and Distributed Infrastructure

In his opening remarks, CIAT Executive Director Robert stated that Taiwan is at a pivotal moment driven by AI-led growth, with AI now serving as a key engine for economic development and capital markets. Amid challenges such as AI sovereignty, cybersecurity, and privacy, building an island-wide AI infrastructure for Taiwan has become a critical industry priority. The government has already launched ten major AI initiatives aimed at creating an island-wide interconnected AI infrastructure, linking distributed AI computing resources through an all-photonics network. Compared with building ultra-large centralized data centers on the scale of 1GW to 5GW, Taiwan and Japan are better positioned to develop distributed AI infrastructure interconnected via all-photonics networks—an approach highly aligned with IOWN's development direction across Asia.

IOWN Global Forum Board Director Rong-Ruey Lee noted that the IOWN Global Forum now has approximately 180 member companies, 12 of which are based in Taiwan. He further emphasized that building distributed AI data centers through end-to-end APN, and progressively realizing the AI Computing Continuum, represents a key vision on the path toward 2030.

Keynote Presentations and Industry Insights

NTT: IOWN, Photonics-Powered AI Infrastructure

NTT noted that the world is entering the era of AI inference. Compared with centralized training, inference places greater emphasis on real-time responsiveness, distributed deployment, and energy efficiency—demanding an entirely new approach to infrastructure. Because optical transmission offers advantages in low power consumption and low heat generation, IOWN leverages all-photonics networks combined with optical-electronic convergence technology to build next-generation AI infrastructure spanning Scale-up, Scale-out, and Scale-across architectures.

To validate this vision, NTT shared several real-world application results. In an AI remote robotic control case, a zero-carbon data center in Hokkaido used IOWN's APN to control autonomous mobile robots at a logistics center in Chiba in real time, across a distance of approximately 800 kilometers—demonstrating that computing resources can overcome geographic and power constraints. In a cloud endoscopy application, medical imaging can be uploaded and analyzed instantly to support remote diagnosis, significantly improving healthcare efficiency. Additionally, in a cross-border scenario, the "Taiwan AI Impact" demonstration in April 2026 successfully connected distributed data centers in Taiwan and Japan, 3,240 kilometers apart, completing real-time cross-border video analysis with a latency of just 19 milliseconds. With this achievement, NTT declared that IOWN has officially moved from technology validation into practical market deployment.

1Finity: Powering AI Across Scales with All-Photonics Network

1Finity, a wholly owned subsidiary established after Fujitsu consolidated its network business in 2025, brings more than 90 years of experience in networking and optical communications technology. 1Finity observed that as AI inference increasingly moves toward the end-user, data center and network architectures are shifting from centralized to distributed models. Rather than relying on long-distance power transmission to support massive AI computing demands, a more efficient approach is to deploy training data centers in regions with abundant power supply, deploy inference data centers closer to end users, and interconnect them via an all-photonics network (APN).

1Finity also shared a field validation case from Kyushu, Japan, in which APN was used to connect different data centers for distributed GPU large-model inference. After adopting RDMA, data transmission time was significantly reduced compared with traditional TCP/IP. The company also demonstrated a multi-data-center interconnection architecture, using optical networks to link data centers across different regions to realize the concept of "inference near the user, training near the power source." 1Finity emphasized that all-photonics network technology is becoming a cornerstone of AI infrastructure, helping distributed AI architectures balance performance, energy efficiency, and operational resilience.

GigaIO: Your Data Is Generated at the Edge — The Compute Is Finally Following

GigaIO approached the topic from the perspective of edge computing, noting that AI inference is driving a shift in system architecture thinking. According to data cited by the company, approximately 75% of enterprise data today is generated at the edge, yet as much as 90% of it is never effectively utilized—not because the data lacks value, but because of challenges related to transmission, cost, and processing efficiency. As sensors continue to generate ever-larger volumes of data, deploying compute capacity at the point of data generation has become a more efficient choice.

Among the various computing platforms on display, GigaIO showcased an AI computing system built into a suitcase form factor, designed for rapid deployment across different locations—demonstrating the ability to bring compute power directly to where the data resides. GigaIO also shared a case involving the U.S. Army performing facial and voice recognition tasks in remote, disconnected environments, illustrating the practical value edge AI computing can deliver in scenarios where cloud connectivity cannot be relied upon.

Saviah Technologies: Private 5G as the Last Mile for AI Edge — Why the Core Must Be Open and Programmable

Saviah Technologies focused on the wireless and software layers of the IOWN architecture, noting that private 5G is increasingly becoming a critical foundation for edge AI deployment, offering low-latency, secure, and localized data processing capabilities. However, many current deployments remain constrained by closed and inflexible core network architectures, making it difficult to meet the customization and dynamic scheduling requirements of AI workloads.

To address this, Saviah Technologies proposed an open and programmable 5G core architecture that integrates network, compute, and data resources into a single system. When combined with the deterministic transmission capabilities provided by IOWN's APN, this approach not only extends high-performance connectivity to the last mile but also delivers end-to-end performance guarantees—building infrastructure for large-scale edge AI applications that is secure, controllable, and optimized for performance.

Pegatron: IOWN APN-T 400G Muxponder for Cloud Edge Computing

Pegatron noted that AI model size grows approximately 750-fold every two years, widening the gap between "compute" and "network" capacity by more than three orders of magnitude. To keep up with AI training and inference distributed across locations, Scale-up and Scale-out within a single data center are no longer sufficient—cross-data-center "Scale-across" has become essential. In terms of products and applications, Pegatron demonstrated its 400G OLS/DCI system and presented two real-world deployment scenarios: pairing the APN-T 400G Muxponder with ROADM to connect edge data centers and O-RAN sites, integrating regional small and mid-sized data centers into a single distributed AI network; and supporting the IOWN Data Center Exchange (DCX) service demonstrated by NTT, Orange, and Telefónica at OFC, enabling multiple data centers to interconnect flexibly in a "many-to-many" configuration. Pegatron emphasized that IOWN's APN is central to this transformation in data center interconnects, supporting the path toward the 2030 vision—enabling geographically dispersed compute resources to operate collaboratively as if they were within a single data center.

Chunghwa Telecom: IOWN APN-Driven AI Use Cases

Chunghwa Telecom approached the topic from an application perspective, noting that the rapid development of AI is driving deep convergence between computing and networking. By integrating computing cloud and APN network management, distributed computing resources can collaborate across clouds, improving both computational efficiency and overall system resilience.

Citing the 2026 "Taiwan AI Impact" smart city demonstration—developed jointly with NTT, the National Center for High-performance Computing (NCHC), and Accton/Edgecore—Chunghwa Telecom connected sites in Taipei, Tainan, and Tokyo through domestic APN and international IOWN APN networks, showcasing multiple cross-regional application scenarios. Even with compute resources distributed across different locations, street-view image recognition could be completed at near-local processing speed. In the event of an anomaly at the primary data center, inference models could be quickly migrated to a remote site to take over operations, ensuring uninterrupted service. The system also integrates 5G Edge AI and connected-vehicle technology to provide real-time safety alerts at unsignalized intersections, and in smart healthcare scenarios, enables rapid transmission of large medical images so that healthcare professionals can access diagnostic data in real time. Chunghwa Telecom stated that, from smart transportation to smart healthcare, IOWN's APN is progressively bringing real-time cross-domain collaboration, highly resilient infrastructure, and intelligent applications into reality.

Edgecore: Taiwan AI Impact — IOWN End-to-End DCI to APN Deployment

Edgecore showcased how an end-to-end IOWN system operates in real-world conditions through its "Taiwan AI Impact" smart traffic monitoring demo. On site, traffic footage was captured via smart poles at intersections and transmitted over a 25-kilometer APN optical link built by the National Center for High-performance Computing (NCHC), connecting IOWN cabinets in the Southern Taiwan Science Park and Shalun for GPU inference. The cabinets integrate key components including APN-G, OWS, APN-T, CSR440, data center networking, SONiC switches, and CDI dynamic resource pooling, forming a complete end-to-end infrastructure. Test results showed that regardless of which site processed the video feed, traffic analysis remained consistently "smooth," with inference times stable between 1.3 and 1.5 seconds, largely unaffected by distance. From the application's perspective, the two data centers—25 kilometers apart—function as if they were a single local facility, sharing distributed computing resources in real time. Through its open networking platform, data center switching infrastructure, SONiC network solutions, and optical networking equipment, Edgecore has advanced IOWN from a conceptual architecture into a deployable, manageable solution that can integrate with existing data center and telecom environments.

UfiSpace: The Photonics Revolution — 1.6T DCI Performance in an IOWN Framework

UfiSpace noted that as AI drives rapid growth in network traffic, data center interconnects are moving toward the 800G and even 1.6T era. Traditional electronic switching architectures are approaching their limits in terms of power consumption and latency, making APN-based all-photonics networks a key direction for the next stage of development.

UfiSpace presented two real-world IOWN case studies developed in collaboration with NTT. In a sports broadcasting scenario, 400G IP-over-DWDM technology enabled real-time transmission of uncompressed 4K/8K video over distances exceeding 80 kilometers, maintaining extremely low latency and jitter so that on-site camera crews and remote production centers could collaborate as if working in the same location. A second case focused on data center interconnects, using an open, disaggregated 400G network architecture to connect multiple data centers—improving transmission efficiency while reducing deployment costs, operating expenses, and overall energy consumption. UfiSpace stated that as AI clusters continue to scale, all-photonics networks will become a critical foundation for future AI infrastructure, helping Scale-up, Scale-out, and Scale-across architectures achieve a balance among high throughput, low latency, energy efficiency, and data security.

Conclusion: The Era of Distributed AI Infrastructure Has Arrived

The industry shares a common vision for next-generation AI infrastructure: connecting distributed computing resources through all-photonics networks to build a distributed AI architecture that combines high performance, low latency, high resilience, and energy efficiency. As AI applications rapidly proliferate, data center competition is no longer confined to individual sites—the real challenge lies in enabling cross-regional, cross-cloud computing resources to operate collaboratively as though they were part of a single data center. This event drew more than 200 industry professionals, reflecting strong industry interest in all-photonics networks and distributed AI infrastructure. Through forum discussions, technology demonstrations, and real-world use cases, attendees witnessed distributed AI data centers moving progressively from concept to practical implementation. With its strong ICT industry foundation and comprehensive supply chain advantages, Taiwan is well positioned at this pivotal intersection of AI and optical communications development. CIAT will continue to work with IOWN ecosystem partners to advance the shared vision of the AI Computing Continuum, moving toward a new era of AI infrastructure.

Demo Solution Showcases